Generative AI promises clear gains across product discovery-to-delivery workflows, but benefits depend on strategy alignment, role redesign, data readiness, and strong guardrails; without these, quality, bias, and IP risks can offset value.

Main Finding

This paper synthesizes evidence that generative AI (GenAI) can meaningfully reshape software product management (SPM) across the ISPMA lifecycle—improving speed, quality, and customer experience—if deployed with explicit strategy alignment, process integration, role redesign, data readiness, and governance/ethical guardrails. The primary contribution is a conceptual framework that maps concrete GenAI capabilities to ISPMA domains and couples operational adoption guidance with governance and ethics controls for responsible scaling.

Key Points

• Scope and contribution
  • Systematic literature review mapping GenAI use-cases to ISPMA SPM domains (strategy → monitoring).
  • Produces a three-layer framework: (1) SPM Lifecycle (ISPMA), (2) GenAI capabilities, (3) governance/ethical controls.
• Major GenAI use-cases in SPM
  • Market research & strategy: synthetic customers, trend analysis, faster idea validation and customer-segmentation discovery.
  • Customer insights & support: conversational agents and analytics that increase agent productivity and personalize responses.
  • Requirements engineering: automated conflict detection in SRS, auto-generation of user stories from epics.
  • Agile planning: AI-based story-point estimation and sprint planning (e.g., GPT2SP).
  • Development & engineering productivity: automated code generation (e.g., Copilot, Codex), comment/doc generation, faster task completion.
  • UI/UX: prototype generation from text prompts, automated usability feedback analysis for rapid iteration.
• Reported outcomes (selected empirical results)
  • Productivity: GitHub Copilot users completed programming tasks ~55.8% faster in an experiment.
  • Estimation accuracy: GPT2SP improved within-project estimates by ~34–57% and cross-project by ~39–49% versus baselines; improved Deep-SE by 6–47%.
  • Support center: deploying a GenAI assistant across 5,179 agents increased worker productivity, improved customer sentiment, and lowered turnover; benefits concentrated among newer/less-skilled staff.
  • Requirement conflict detection: transformer-based pipeline improved F1 by ~4–5% on benchmark datasets.
  • Market forecast & survey: projected generative AI market ≈ $109B by 2030; McKinsey survey reported 67% of respondents saw revenue increases and 79% saw cost decreases from AI adoption.
• Risks & ethical concerns
  • Hallucinations and factual errors in LLM outputs; need for human validation.
  • Black-box models → explainability and trust limitations.
  • Legal/IP exposure for AI-generated content; unclear ownership and infringement risk.
  • Privacy risks (data leakage via prompts), regulatory constraints (GDPR), bias replication, potential for disinformation/deepfakes.
  • Workforce effects: augmentation for many roles but potential displacement and skill-shift requirements.
• Adoption prerequisites & governance
  • Strategy alignment (where GenAI creates value), role redesign (AI-human workflows), process integration (tooling, pipelines), data readiness (quality, lineage), and risk controls (contracts, prompt/data policies, human-in-loop checks).
  • Recommended governance: explicit vendor/customer contract clauses, prompt confidentiality rules, audit trails, bias/privacy impact assessments, and continuous monitoring.

Data & Methods

• Method: Systematic Literature Review (SLR) following Kitchenham-style guidelines; synthesis structured through ISPMA and framed with change/ethics models (McKinsey 7-S, Lewin’s Change Model, Responsible Innovation, GDPR).
• Search strategy: five digital libraries queried (IEEE Xplore, ACM DL, Google Scholar, EBSCOhost, ProQuest Central) using a PIC-style search string for GenAI/LLMs and SPM-related terms.
• Time window and selection: studies published 2019–2023; inclusion limited to peer-reviewed conference/journal/workshop papers in English with full text.
• Screening & corpus: PRISMA-style filtering produced 78 primary studies for analysis.
• Extraction & synthesis: captured publication venue, year, research method, GenAI technology, SPM activity, and key findings; thematic coding mapped evidence to ISPMA domains and governance themes.

Implications for AI Economics

• Productivity and output growth
  • Empirical task-level productivity gains (e.g., Copilot ~56% faster) imply substantial firm-level efficiency improvements if scaled, with the potential to reduce development costs and accelerate time-to-market—feeding into higher firm profits and aggregate GDP effects.
  • Gains are heterogeneous: largest for less-experienced workers and routine tasks, implying short-run reallocation of labor productivity rather than uniform across skill groups.
• Labor market and wage effects
  • Complementarity and substitution: GenAI augments many roles (support agents, junior engineers) and may substitute some task segments—expect skill-biased reallocation, increased demand for AI-proficient roles (prompt engineering, oversight), and potential downward pressure on wages for automatable tasks.
  • Transition costs: retraining, redeployment, and temporary unemployment risk create frictions and require investment in human capital.
• Firm strategy, competition, and market structure
  • Firms that rapidly integrate GenAI into core product management (strategy, discovery, delivery) can capture outsized first-mover advantages through faster iteration and lower marginal costs—potentially increasing concentration in software industries.
  • Platform and vendor lock-in risks as proprietary GenAI tools and models become embedded in development workflows; intellectual property disputes over model training data and outputs can change value capture dynamics.
• Investment and capital allocation
  • Increased returns to investments in data infrastructure, labeling, and governance (privacy, IP protections) as prerequisites to safely realize GenAI benefits.
  • Budget reallocation from routine operational headcount toward AI tooling, oversight, and quality assurance.
• Pricing, monetization, and productization
  • New monetization models: AI-enhanced SaaS tiers, per-use model inference fees, and outcome-based pricing tied to productivity improvements.
  • Need to internalize costs of governance (compliance, auditing) into product pricing and ROI calculations.
• Regulatory and externality considerations with economic consequences
  • IP uncertainty, liability for AI-generated defects or misinformation, and privacy regulation (e.g., GDPR) can raise compliance costs and slow adoption—policy choices will materially affect diffusion and economic returns.
  • Negative externalities (misinformation, biased outputs) impose social costs and potential reputational/financial risks for firms; internalizing these costs (through audits, insurance, regulation) will change the economic calculus of deployment.
• Evidence gaps relevant to economics
  • Limited causal and longitudinal studies quantifying firm-level ROI, labor reallocation effects, macroeconomic impacts, and distributional consequences across sectors.
  • Need for standardized metrics to value quality improvements (customer experience, defect reduction) and to model long-run equilibrium effects on wages, employment, and industrial structure.

Suggested immediate actions for economists and decision-makers - Firms: pilot GenAI in high-frequency, well-scoped SPM tasks (e.g., estimation, support) with measurement frameworks for productivity, quality, and customer impact; invest in data and governance before scaling. - Researchers: prioritize causal field experiments, longitudinal workforce studies, and cost–benefit analyses that quantify net economic impacts and distributional outcomes. - Policy makers: clarify IP rules for AI outputs, set standards for prompt/data confidentiality, and design transition-support policies (retraining, income buffers) to manage labor-market shifts.

If you want, I can convert the framework figure and table summaries into a one-page checklist for product leaders or draft a short list of research designs that would address the key evidence gaps for economic impact assessment.